Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Indag. Mathem., N.S., 20 (1), 87–100 March, 2009 Generalized balancing numbers ✩ by Kálmán Liptai a , Florian Luca b , Ákos Pintér c and László Szalay d a Institute of Mathematics and Informatics, Eszterházy Károly College, Leányka út 4, H-3300 Eger, Hungary b Instituto de Matemáticas, UNAM, Campus Morelia, Ap. Postal 61-3 (Xangari), CP 58089, Morelia, Michoacán, Mexico c Institute of Mathematics, University of Debrecen, Egyetem tér 1, H-4010 Debrecen, Hungary d Department of Mathematics and Statistics, University of West Hungary, Erzsébet utca 9, H-9400 Sopron, Hungary Communicated by Prof. R. Tijdeman ABSTRACT The positive integer x is a (k, l)-balancing number for y (x y − 2) if 1k + 2k + · · · + (x − 1)k = (x + 1)l + · · · + (y − 1)l , for fixed positive integers k and l . In this paper, we prove some effective and ineffective finiteness statements for the balancing numbers, using certain Baker-type Diophantine results and Bilu–Tichy theorem, respectively. 1. INTRODUCTION Let y, k and l be fixed positive integers with y 4. We call the positive integer x ( y − 2) a (k, l)-power numerical center for y , or a (k, l)-balancing number for y if (1) 1k + · · · + (x − 1)k = (x + 1)l + · · · + (y − 1)l . ✩ K. Liptai was supported by the Hungarian National Foundation for Scientific Research Grant No. T 048945 MAT. F. Luca was supported by Grant SEP-CONACyT 46755 and a Guggenheim Fellowship. Á. Pintér was supported by the Hungarian National Foundation for Scientific Research Grants Nos. K 75566, K 067580 and by the Bolyai Scholarship. L. Szalay was supported by the Hungarian National Foundation for Scientific Research Grant No. T 048945 and by the Eötvös Scholarship of Hungary. E-mails: [email protected] (K. Liptai), fl[email protected] (F. Luca), [email protected] (Á. Pintér), [email protected] (L. Szalay). 87 The special case k = l of this definition is due to Finkelstein [9] who proved that infinitely many integers y possess (1, 1)-power centers (see also [2]), and that there is no integer y > 1 with a (2, 2)-power numerical center. The proofs depend on the theory of Pell equations and the resolution of the Thue equations X 3 + 2Y 3 = 11 and 33, respectively, in integers X, Y . We note that the particular case k = l = 1 is strongly related to another problem called the house problem (see, for example, [1]). In his paper, Finkelstein conjectured that if k > 1, then there is no integer y > 1 with a (k, k)-power numerical center. Later, using a result of Ljunggren [11] and Cassels [7] on triangular numbers whose squares are also triangular, Finkelstein [14] confirmed his own conjecture for k = 3. Recently, Ingram [10] proved Finkelstein’s conjecture for k = 5. In this paper, we prove a general result about (k, l)-balancing numbers. Unfortunately, we cannot deal with Finkelstein’s conjecture in its full generality. However, we obtain the following theorem. Theorem 1. For any fixed positive integer k > 1, there are only finitely many positive pairs of integers (y, l) such that y possesses a (k, l)-power numerical center. The case k = l has already been dealt with recently by Ingram [10] with a method similar to ours. The proof splits naturally in two cases. The first case is when 1 l k . Since k is fixed and there are only finitely any such l , we may assume that l is also fixed. Our proof now uses an ineffective statement of Rakaczki [13]. The second case is when k < l , and here we show by Runge’s method that there are no positive integers y possessing a (k, l) numerical center. Because of the use of the result from [13], our Theorem 1 is ineffective in case l k in the sense that we cannot provide an upper bound for possible numerical centers x in terms of k unless l = 1 or l = 3. In these cases, we have the following theorem. Theorem 2. Let k be a fixed positive integer with k 1 and l ∈ {1, 3}. If (k, l) = (1, 1), then there are only finitely many (k, l)-balancing numbers, and these balancing numbers are bounded by an effectively computable constant depending only on k . We note that some numerical centers do exist. For example, in the case (k, l) = (2, 1), we can rewrite equation (1) as 2x 3 + 4x = 3y 2 − 3y, which is an elliptic curve whose short Weierstrass normal form is u3 +72u+81 = v 2 (via the bi-rational transformation v = 18y − 9 and u = 6x ). Using the program package MAGMA, we solved this elliptic equation and its solutions lead to three (2, 1)-balancing numbers x , namely 5, 13 and 36. 88 2. AUXILIARY RESULTS For the proofs of our results, we need some auxiliary results. For an integer k 1, we write Sk (x) = 1k + 2k + · · · + (x − 1)k . These expressions are strongly related to the Bernoulli polynomials. In the next lemma, we summarize some of the well-known properties of Bernoulli polynomials. For the proofs, we refer to [12]. Lemma 1. Let Bn (X) denote the nth Bernoulli polynomial and put Bn = Bn (0) for n = 1, 2, . . . . Further, let Dn be the denominator of Bn . We then have: (A) (B) (C) (D) (E) (F) 1 Sk (X) = k+1 (Bk+1 (X) − Bk+1 ); Bn (X) = X n + nk=1 nk Bk X n−k ; Bn (X) = (−1)n Bn (1 − X); B1 = − 12 and B2n+1 = 0 for n 1; (von Staudt–Clausen) D2n = p−1|2n,p prime p; 0 and 1 are double zeros of Sk (X) for odd values of k 3. Further, 0 and 1 are simple zeros of Sk (X) for even values of k . We shall also need the following lemma (for some results of a similar flavor, see the Appendix to [3]). Lemma 2. Let p be prime. Assume that the sum of the digits of n in base p is p . Then there exists an even positive integer k < n such that p divides the denominator of the rational number n Bk k when written in reduced form. Proof. Let first p be odd. Then n = n 1 p α1 + · · · + nt p αt , here 0 α1 < · · · < αt and n1 , . . . , nt ∈ {1, . . . , p − 1}. We now select non-negative integers mi for i = 1, . . . , t , such that mi ni and ti=1 mi = p − 1. It is clear that t this can be done since i=1 ni p . We put k= t mi p αi . i=1 Then, k < n. Further, reducing k modulo p − 1, we get that k ≡ ti=1 mi (mod p − 1), therefore (p − 1) | k . In particular, k is even and p divides the denominator of Bk . 89 Finally, by a well-known theorem of Lucas (see [8], p. 271, items 76 and 77), we have t ni n ≡ k mi (mod p), i=1 therefore p does not divide nk , which completes the proof of the lemma in this case. If p = 2, then n = 2α1 + 2α2 + · · · + 2αt with 0 α1 < α2 < · · · < αt and t 2. Then, taking k = 2α2 , we see that k < n, k is even, and, by Lucas’s Theorem, nk is odd. Since 2 divides the denominator of Bk , we get that 2 divides the denominator of nk Bk . 2 The next lemma is based on a recent deep theorem of Bilu and Tichy [4], as well as on the indecomposability of the Bernoulli polynomials proved by Bilu et al. in [3]. To present Lemma 3, we define special pairs (l, g(X)) as follows. In the sequel, we let δ(X) ∈ Q[X] be a linear polynomial, and q(X) ∈ Q[X] be a non-zero polynomial. Further, for l odd, Sl (X) can be written in the form φl ((X − 1/2)2 ) with some appropriate polynomial φl (X) ∈ Q[X], see [13]. We now define special pairs (l, g(X)) as follows: Special pair of type I: (l, Sl (q(X))), where q(X) is not constant. Special pair of type II: l is odd and g(X) = φl (δ(X)q(X)2 ). Special pair of type III: l is odd and g(X) = φl (cδ(X)t ), where c ∈ Q \ {0} and t 3 is an odd integer. Special pair of type IV : l is odd and g(X) = φl ((aδ(X)2 + b)q(X)2 ), where a, b ∈ Q \ {0}. Special pair of type V : l is odd and g(X) = φl (q(X)2 ). Special pair of type VI: l = 3 and g(X) = δ(X)q(X)2 . Special pair of type VII: l = 3 and g(X) = q(X)2 . Lemma 3. Let l be a positive integer and g(X) ∈ Q[X] be a polynomial of degree greater than 2. Then the equation Sl (x) = g(y) has only finitely many integer solutions x and y , unless (l, g(X)) is a special pair. Proof. This is Theorem 1 in [13]. 2 We now rewrite equation (1) using the polynomials Sk (x) and Sl (y) as the Diophantine equation (2) 90 Sk (x) + Sl (x + 1) = Sl (y). Lemma 4. Assume that k < l and put (3) P (X) = (X + 1)l − Sk (X) ∈ Q[X]. Put x0 for the largest real root of P (X). If x and y is an integer solution to the Diophantine equation (2), then x x0 . In particular, if either P (X) has no real root, or x0 < 2, then the Diophantine equation (2) has no integer solutions x 2 and y x + 2. Proof. Suppose that the integers x 2 and y x + 2 satisfy (2). Since k < l , it follows that the leading coefficient of P (X) is positive and deg(P ) = l . Clearly, if P (X) does not possess a real root, or if x0 < 2, or if 2 x0 < x , then P (x) = (x + 1)l − Sk (x) > 0. Then (4) −x l − Sk (x) < 0 < (x + 1)l − Sk (x). Increasing both sides of the inequality (4) by x l + Sk (x) + Sl (x), we get (5) Sl (x) < x l + Sk (x) + Sl (x) < (x + 1)l + x l + Sl (x), which leads to (6) Sl (x) < Sk (x) + Sl (x + 1) < Sl (x + 2). Since (x, y) is a integer solution of (2), we get that Sk (x) + Sl (x + 1) can be replaced in (6) by Sl (y). Thus, (7) Sl (x) < Sl (y) < Sl (x + 2). The properties of the polynomial Sl together with inequalities (7) imply that y = x + 1. Thus, by (2), Sk (x) = 0, which is impossible. 2 The following three results yield information on the structure of zeros of certain polynomials. Lemma 5. Let p(X) = an X n + · · · + a1 X + a0 be a polynomial with integral coefficients for which a0 is odd, 4 | ai for all i = 1, . . . , n, and ord2 (an ) = 3. Then every zero of p is simple. Proof. This is Lemma 4 in [6]. 2 One of the most surprising theorem on zeros of certain polynomials related to Sk (x) is due to Voorhoeve, Győry and Tijdeman [15]. They proved if k ∈ / {1, 3, 5} then the polynomial Sk (x) + R(x) possesses at least three zeros of odd multiplicities for every polynomial R(x) with rational integer coefficients. Of course, this general statement is not true for some polynomials with rational coefficients, however, we obtain Lemmas 6 and 7. 91 Lemma 6. Each one of the polynomials F1 (X) = 8Sk (X) + (2X + 1)2 , k > 1, has at least three zeros of odd multiplicities. Proof. Let d be the smallest positive integer for which 8d Bk+1 (X) − Bk+1 ∈ Z[X]. Then, by (D) and (E) of Lemma 1, d is an odd square-free integer. We now show that the polynomial P1 (X) = 8d Bk+1 (X) − Bk+1 + d(k + 1)(2X + 1)2 has at least three zeros of odd multiplicities. Note that the leading coefficient of P1 equals 8d . Since d is odd and is the smallest positive integer such that 8d(Bk+1 (X)− Bk+1 ) ∈ Z[X], it follows that the content of P1 (X) (i.e. the greatest common divisor of all its coefficients) is a power of 2 dividing 8. If k is even, the fact that P1 (X) has at least three simple zeros is a simple consequence of Lemma 5. Assume now that k is odd. If P1 (X) is associated to a complete square in Q[X], then we get that (8) P1 (X) = aR(X)2 , where a is an integer and R(X) ∈ Z[X] is a polynomial with positive leading term. By writing a = a1 r 2 , with integers a1 and r such that a1 is square-free, and by replacing R(X) by rR(X), we may assume that a is square-free. It is clear that a > 0. We also assume that k 5, since the case k = 3 can be checked by hand. If 2 k + 1, it is then easy to see that the content of P1 (X) is 2 (all the coefficients of P1 (X) are even, and the last is d(k + 1), therefore it is not a multiple of 4). Hence, by Gauss Lemma, a = 2 and the content of R(X) is 1. Writing R(X) = a0 X (k+1)/2 + a1 X (k−1)/2 + · · · + a(k+1)/2 , and identifying the first three coefficients in (8), we get 8d = aa02 , −4d(k + 1) = 2aa0 a1 , 2dk(k + 1) = a a12 + 2a0 a2 . 3 The first relation above forces a0 = 2 and d = 1. The third relation above shows that a1 is even, therefore 2aa0 a1 is a multiple of 16. Now the second relation above contradicts the fact that 2 k + 1. If 4 | k + 1, then the content of P1 (X) is 4, unless (see Lemma 2) k + 1 is a power of 2, in which case it is 8. Thus, a = 1, unless k + 1 is a power of 2, in which case a = 2. Identifying leading terms in (8), we get that 8d = aa02 , therefore d = 1 92 and a = 2. Thus, k + 1 = 2α , for some α 3. However, since d = 1, it follows, by Lemma 2, that the sum of the digits of k + 1 in base 3 is at most 2. Thus, since k + 1 is not a power of 3 (because k + 1 is even), we get that k + 1 = 3β + 3γ for some 0 β γ . Hence, 2α = 3β + 3γ , therefore β = 0. Since the largest solution of the Diophantine equation 2α = 1 + 3γ is α = 2, γ = 1, we get k + 1 = 4, therefore k = 3, which is a contradiction. We now have to exclude the remaining case in which (9) P1 (X) = aX 2 + bX + c R 2 (X), where both aX 2 + bX + c and R(X) are in Z[X], such that aX 2 + bX + c has two distinct zeros. Up to replacing R(X) by −R(X), we may assume that the leading coefficient of R(X) is positive. Further, because the content of P1 (X) is a power of 2 dividing 8, it follows that gcd(a, b, c) is a power of 2. By writing gcd(a, b, c) = 2α for some non-negative integer α , and replacing R(X) by 2α/2 R(X), we may assume that α = 0, or 1. Hence, gcd(a, b, c) is either 1 or 2. If k + 1 is even but not divisible by 4, then P1 (X)/2 is a polynomial in Z[X] having odd constant term and all other coefficients even. Thus, P1 (X)/2 ≡ 1 (mod 2). Hence, it can be factored as 2 P1 (X)/2 = 2S1 (X) + 1 2S2 (X) + 1 , with some polynomials Si (X) ∈ Z[X] for i = 1, 2. However, the leading coefficient of P1 (X)/2 is not divisible by 8. Assume now that 4 | k + 1. When k = 3, one can check by hand that P1 (X) does not have the form shown at (9). Assume now that k + 1 8. Note that the content of P1 (X) is 4, unless k + 1 is power of 2, when it is 8. It now follows that R1 (X) = R(X)/2 ∈ Z[X]. Thus, (10) P1 (X)/4 = aX 2 + bX + c R1 (X)2 . We now write R1 (X) = a0 X (k−1)/2 + a1 X (k−1)/2−1 + a2 X (k−1)/2−2 + · · · + a(k−1)/2 . Identifying the first three coefficients in P1 (X)/4, we get, on the one hand the polynomial P1 (X)/4 = 2dX k+1 − d(k + 1)X k + dk(k + 1) k−1 + ···, X 6 while on the other hand the polynomial aX 2 + bX + c a02 X k−1 + 2a0 a1 X k−2 + a12 + 2a0 a2 X k−3 + · · · = aa02 X k+1 + ba02 + 2aa0 a1 X k + ca02 + 2ba0 a1 + a a12 + 2a0 a2 X k−1 + · · · 93 which leads to the relations aa02 = 2d, ba02 + 2aa0 a1 = −d(k + 1) and dk(k + 1) ca02 + 2ba0 a1 + a a12 + 2a0 a2 = . 6 The first relation above shows that a0 = 1, a = 2d . The second one now becomes (11) b = −d(k + 1 + 4a1 ), while the third one now reads (12) d(k(k + 1) c = d 2(k + 1)a1 + 6a12 − 4a2 + . 6 Clearly, d | a and the above relations (11) and (12) for b and c , show that d | b , and if there exists a prime p > 3 such that p | d , then p | gcd(a, b, c). Since gcd(a, b, c) ∈ {1, 2}, we get that d = 1 or d = 3. To rule out the possibility that d = 3, assume that 3 | k + 2. Then the above formula (12) for c shows that d = 3. Identifying the last coefficient in (9) and using the fact that Sk (0) = 0 (by (F) of Lemma 1), we get (13) d(k + 1) = c(2a(k−1)/2 )2 , and since 3 divides d but not k + 1, we get that 3 | c . Since d | a and d | b , we obtain 3 | gcd(a, b, c), which is again a contradiction. Thus, 3 does not divide k + 2, therefore 3 | k(k + 1). Now relation (12), shows again that d | c . Hence, d | gcd(a, b, c), which implies that d = 1. Lemma 2 implies now that the sum of the digits of k + 1 is base 3 is 2. Since 4 | k + 1, we get that k + 1 cannot be a power of 3, therefore k + 1 = 3α0 + 3α1 for some 0 α0 α1 . Since 4 | k + 1, and k + 1 = 3α0 (3α1 −α0 + 1), we get that 4 | 3α1 −α0 + 1. This shows that α1 − α0 is odd. Further, since for an odd positive integer s , we have that 4 3s + 1, we get that (k + 1)/4 is odd. Clearly, 2 a and relation (12) show that c is even. Now relation (13) together with the facts that c is even and 4 k + 1 leads to a contradiction. 2 Lemma 7. Each one of the polynomials F2 (X) = 4Sk (X) + X 2 (X + 1)2 , k 1, k = 3, has at least three zeros of odd multiplicities. Proof. We follow the same method as in the proof of Lemma 6. We use again d for the least positive integer such that 4d(Bk+1 (X) − Bk+1 ) ∈ Z[X]. By (D) and (E) of Lemma 1, we have that d is odd and square-free. 94 For the polynomial F2 (X), we have P2 (X) = 4d Bk+1 (X) − Bk+1 + d(k + 1)X 2 (X + 1)2 ∈ Z[X]. Assume first that k + 1 is odd. We then have to exclude the case when P2 (X) = (aX + b)R 2 (X), where aX + b and R(X) ∈ Z[X]. Since d is square-free and odd, further 0 is a simple zero of P2 (X) (by (F) of Lemma 1), we obtain P2 (X) = aXR 2 (X). The coefficient of X in R 2 (X) is even, thus the coefficient of X 2 in P2 (X) is also even, which is a contradiction. Assume now that k + 1 > 4 is even. Then, we have to exclude that either P2 (X) = aR(X)2 , or P2 (X) = (aX 2 + bX + c)R(X)2 with some polynomial R(X) ∈ Z[X], and some integers a, b and c , with a = 0. We first look at the case P2 (X) = aR(X)2 . We may assume again that R(X) has positive leading coefficient and that a > 0 is square-free. The content of P2 (X) is a power of 2, therefore a = 1 or 2. We assume that k + 1 > 8, since the smaller cases can be checked by hand. Writing R(X) = a0 X (k+1)/2 + a1 X (k+1)/2−1 + · · · and identifying the first 5 coefficients we get, on the one hand, that P2 (X) is the polynomial d(k + 1)k k−1 X 3 d(k + 1)k(k − 1)(k − 2) k−3 − X 180 4dX k+1 − 2d(k + 1)X k + (note that k − 3 > 4, therefore the first 5 terms in P2 (X) are the same as the first 5 terms in 4dSk (X)), while on the other hand, we get the polynomial aa02 X k+1 + 2aa0 a1 X k + a a12 + 2a0 a2 X k−1 + a(2a0 a3 + 2a1 a2 )X k−2 + a a22 + 2a0 a4 + 2a1 a3 X k−3 + · · · . Hence, we obtain the relations (14) 4d = aa02 , −2d(k + 1) = 2aa0 a1 , d(k + 1)k = aa12 + 2aa0 a2 , 3 as well as (15) 2aa0 a3 + 2aa1 a2 = 0 and aa22 k + 1 2d + 2aa0 a4 + 2aa1 a3 = − . 4 15 95 The first relation in (14) above together with a ∈ {1, 2}, gives d = 1, a0 = 2, and a = 1. Hence, by Lemma 2, k + 1 is either a power of 3, or of the form 3α + 3β for some 0 α β . Since k + 1 is even, it cannot be a power of 3, therefore k + 1 = 3α (3β−α + 1). If β − α is even, then 2 k + 1, and if β − α is odd, then 4 k + 1. Now, the second relation in (14) above gives a1 = −(k + 1)/2, and the third relation in (14) above together with the fact that k + 1 is even shows that a1 is even. Thus, 4 k + 1, which shows that k+1 is odd. Finally, since a0 and a1 are both even, 4 reducing the second of relations (15) modulo 4 we get a22 ≡ 2 (mod 4), which is the desired contradiction in this case. It remains to deal with the case when (16) P2 (X) = aX 2 + bX + c R(X)2 , where a, b, c are integers and R(X) ∈ Z[X]. As before, we assume that R(X) has positive leading coefficient, that a > 0, and gcd(a, b, c) = 1 or 2. We write R(X) = a0 X (k−1)/2 + a1 X (k−1)/2−1 + · · · + a(k−1)/2 , assume that k 5 and identify the first three coefficients in (16) to get (17) aa02 = 4d, ba02 + 2a0 aa1 = −2d(k + 1), and (18) dk(k + 1) ca02 + 2ba0 a1 + a a12 + 2a0 a2 = . 3 The first relation (17) shows that d | a and then the second relation (17) shows that d | b . Now relation (18) shows that if there exists a prime p > 3 dividing d , then p | gcd(a, b, c), which is a contradiction. Thus, d is a divisor of 3. To rule out the case d = 3, suppose that 3 | k + 2. Then, relation (18) shows that d = 3. Evaluating relation (16) at x = 1 and using (F) of Lemma 1 (for Sk (1) = 0), we get 4d(k + 1) = (a + b + c)R(1)2 . Since 3 | k + 2, we get that 3 does not divide k + 1. Hence, 3 | (a + b + c), and since 3 divides both a and b , we get that it divides also c , which is again a contradiction. Thus, d = 1, and since k + 1 is even, we get that k + 1 = 3α + 3β with 0 α β . If β − α is even, then 2 k + 1 and if β − α is odd, then 4 k + 1. k−3 in P (X) and it is even but If 4 k + 1, then 4 k+1 2 4 B4 is the coefficient of X k+1 not a multiple of 4, while if 2 k + 1, then 4 2 B2 is the coefficient of X k−1 in P2 (X), and is also even but not a multiple of 4. In conclusion, the content of P2 (X) is 2, which means that gcd(a, b, c) = 2. 96 We now return to relations (17) and (18) and note that a0 ∈ {1, 2}. If a0 = 2, then a = 1, which is false. Thus, a0 = 1 and a = 4. Now the second relation (17) shows that 4 | b . Relation (18) shows that 4 | c too, which leads to the contradiction 4 = gcd(a, b, c), unless 2 k + 1. So, let us assume that 2 k + 1. Note that in this case relation (18) implies that c ≡ 2 (mod 4). Now let us note that by (F) of Lemma 1, we have that 0 is a double roots of Sk (X). Thus, 0 is also double root of P2 (X). Hence, a(k−1)/2 = 0. Put F (X) = P2 (X)/X 2 , and R1 (X) = R(X)/X . Identifying the last two coefficients in the equation F (X) = aX 2 + bX + c R1 (X)2 2 , = aX 2 + bX + c · · · + 2a(k−3)/2 a(k−5)/2 X + a(k−3)/2 we get (19) 2k(k + 1)Bk−1 + k + 1 = c(a(k−3)/2 )2 , 2 2(k + 1) = 2ca(k−3)/2 a(k−5)/2 + ba(k−3)/2 . The first relation (19) shows that 2k(k +1)Bk−1 ∈ Z. Since Bk−1 is a rational number whose denominator is an even square-free integer, it follows that 2k(k + 1)Bk is congruent to 2 modulo 4. Since 2 k + 1, we get that the left-hand side of the first equation (19) is a multiple of 4. Since 2 c, we get that a(k−3)/2 is even. We now immediately get that the right-hand side of the second equation (19) is a multiple of 8, whereas its left-hand side is not. This final contradiction concludes the proof of this lemma. 2 Finally, we recall a special case of a result by Brindza [5]. Lemma 8. Let f (X) ∈ Q[X] be a polynomial having at least three zeros of odd multiplicities. Then the equation f (x) = y 2 in integers x and y implies that max(|x|, |y|) < c , where c is an effectively computable constant depending only on the coefficients and degree of f . Proof. See Theorem in [5]. 2 3. PROOFS We start with the proof of Theorem 2 since it will be needed in the proof of Theorem 1. Proof of Theorem 2. Since x(x − 1) S1 (x) = 2 and S3 (x) = x(x − 1) 2 2 , 97 equation (1) leads to the equations 8Sk (x) + (2x + 1)2 = (2y − 1)2 and 2 2 4Sk (x) + x(x + 1) = y(y − 1) , respectively. Now, apart from the case when in the second equation k = 3, the fact that such equations have only finitely many effectively computable integer solutions is a simple consequence of Lemmas 6–8. We recall that Finkelstein resolved the case (k, l) = (3, 3). Thus, our proof is complete. 2 Proof of Theorem 1. We first assume that k = l > 3. If x is a (k, k)-balancing number for y , then, from (1), we have 2Sk (x) + x k = Sk (y). Recall that for odd values of k , Sk (X) = φk ((X − 1/2)2 ) with an appropriate polynomial φk (X) with rational coefficients, and the leading coefficient of Sk (X) and φk (X) is 1/(k + 1). We show that the identities 2Sk (X) + X k = Sk δ(X) and, for odd k , 2Sk (X) + X k = φk q(X) , where δ(X) and q(X) are polynomials with rational coefficients of degree 1 and 2, respectively, are impossible. Indeed, if the leading coefficient of δ(X) or q(X) is a ∈ Q, then the leading coefficient of Sk (δ(X)) or φk (q(X)) is a k+1 /(k + 1) or k+1 a 2 /(k + 1), respectively, which cannot be 2/(k + 1). Thus, (k, 2Sk (X) + X k ) is not a standard pair, and now Lemma 3 completes the proof. We follow a similar approach for k > l 1. By Theorem 2, we may assume that l∈ / {1, 3}. Since k is fixed, and there are only finitely many such l , we may assume that l is also fixed. Then, from (2), we have Sk (x) + Sl (x + 1) = Sl (y). We prove that the identities (20) Sk (X) + Sl (X + 1) = Sl q(X) and, for odd l , (21) 98 Sk (X) + Sl (X + 1) = φl q(X) , where q(X) is a polynomial with rational coefficients, are impossible. Let d and a be the degree and leading coefficient of q(X). On comparing the degrees and the leading coefficients in (20) and (21), we obtain k + 1 = (l + 1)d and 1 ad = , k+1 l+1 and (22) k+1= l+1 d 2 and 1 ad = , k+1 l+1 respectively. From these relations, we get a d = 1/d or 2/d . Since d > 1, we see that a = 1/b, where b is an integer, and bd = d or bd = d/2. The first equation has no integer solutions with d > 1, while the only solution in integers of the second equation with d > 1 is d = 2, b = ±1. However, when d = 2, the first relation (22) shows that k = l , which is not allowed. We now deal with the case l > k . In this case, the fact that equation (2) has no positive integer solutions x and y is a direct consequence of Lemma 4. By that lemma, it is sufficient to show that the polynomial P (X) = (X + 1)l − Sk (X) has no real zero 2. The estimate x Sk (x) = 1 + 2 + · · · + (x − 1) < k k t k dt = k 0 x k+1 xl , k+1 2 provides P (x) = (x + 1)l − Sk (x) > (x + 1)l − xl >0 2 for all x 0. Thus, there is no (k, l)-balancing number with k < l . 2 ACKNOWLEDGEMENTS The authors are grateful to the referee for his/her remarks and suggestions. REFERENCES [1] Adams J.P. – Puzzles for Everybody, Avon Publications, New York, 1955, p. 27. [2] Behera A., Panda G. – On the square roots of triangular numbers, Fibonacci Quart. 37 (1999) 98–105. [3] Bilu Yu., Brindza B., Kirschenhofer P., Pintér Á., Tichy R. – Diophantine equations and Bernoulli polynomials (with an appendix by Schinzel, A.), Compos. Math. 131 (2002) 173–188. [4] Bilu Yu., Tichy R. – The diophantine equation f (x) = g(y), Acta Arith. 95 (2000) 261–288. [5] Brindza B. – On S -integral solutions of the equations y m = f (x), Acta Math. Hungar. 44 (1984) 133–139. [6] Brindza B., Pintér Á. – On equal values of power sums, Acta Arith. 77 (1996) 97–101. [7] Cassels J.W. – Integral points on certain elliptic curves, Proc. London Math. Soc. 3 (1965) 55–57. [8] Dickson L.E. – History of the Theory of Numbers, vol. I, New York, Chelsea, 1952. [9] Finkelstein R.P. – The House problem, Amer. Math. Monthly 72 (1965) 1082–1088. 99 [10] Ingram P. – On k -th power numerical centres, C. R. Math. Acad. Sci. R. Can. 27 (2005) 105–110. [11] Ljunggren W. – Solution complète de quelques équations du sixième degré à deux indéterminées, Arch. Math. Naturv. 48 (1946) 177–211. [12] Rademacher H. – Analytic Number Theory, Springer-Verlag, New York–Heidelberg, 1973. [13] Rakaczki Cs. – On the diophantine equation Sm (x) = g(y), Publ. Math. Debrecen 65 (2004) 439– 460. [14] Steiner R. – On k th-power numerical centers, Fibonacci Quart. 16 (1978) 470–471. [15] Voorhoeve M., Győry K., Tijdeman R. – On the Diophantine equation 1k + 2k + · · · + x k + R(x) = y z , Acta Math. 143 (1979) 1–8. (Received November 2007) 100